only overburden of the layer is changing settling resistance

Run tests with gitlab ci on every commit

See merge request !1

`xr.read_mfdataset` is concating in the order of the file list that is
provided. Glob seems to sort differently on linux and windows.

read default model parameters as globals from a config module.

evolve model state is now using apply_ufunc which makes it possible to
use dask under the hood.

This parameter controls how fast
the transition towards `rho_max_wet` is taking place when delta_swe is
negative. "Wetting speed" is maybe not the best name.

The loop in snowpack_evolution is now tidy and timestep_forward can also
be used by an operational implementation.

start to refactor in order to be able to test parts of the algorithm.

since `rho_max` is changed when SWE decreases, we assume that the whole
snowpack is already wet and accordingly change `rho_max`in all layers at
the same time. No need to move down from the top.